Research On Semi-supervised Classification Algorithms For Hyperspectral Imagery Based On Swarm Intelligent Optimization Algorithm

With the improvement of the resolution of remote sensing sensors,hyperspectral image processing technology has become an important research direction of remote sensing.Hyperspectral remote sensing images have high spectral resolution,which can provide more detailed information about the classification of objects.Image classification is an important content of remote sensing technology research.However,the high-dimensional non-linearity of the hyperspectral image data,the high inter-band correlation,and the limited set of training samples have posed a great challenge to the classification work.Therefore,how to quickly and accurately classify the objects of hyperspectral remote sensing images is becoming one of the hotspot problems in hyperspectral imagery processing.Semi-supervised algorithm,combining the advantages of supervised algorithm and unsupervised algorithm,can improve the classification accuracy by using the information of unlabeled samples,and become the preferred solution for hyperspectral image classification.In view of the above problems,we study some typical semi-supervised algorithms,and two new semi-supervised learning algorithms are proposed in this paper.The details are as follows:1.A classification framework of active learning and semi-supervised learning is proposed.The proposed algorithm takes into account the difference and complementarity between active learning and semi-supervised learning.Combined with the pseudo-labeling technique based on iterative verification and the improved DE algorithm,the training sample set with high information content and high confidence can be selected to update the training sample set,and then the classification model gets a higher accuracy.Experimental results show that the proposed algorithm can effectively utilize the information of unlabeled samples,and reduce the bad influence of mislabeled samples in the learning process.2.Hyperspectral semi-supervised classification algorithm based on spatial spectrum information is proposed.Firstly,the spatial information,extracted by Gabor filter,is stacked with the spectral information.Then the training sample set is extended from two aspects: firstly,the neighborhood samples of the high confidence sample are screened;then the most valuable samples are selected combining MS with TLBO algorithm.After that,according to the new sample set,SVM based on spatial information is trained and used to classify the test samples.Finally,the image noise points can be removed by smoothing the center sample points with neighborhood information.Experimental results show that the proposed algorithm can improve the accuracy of the classifier more effectively by using the spatial information.